I am just curious to what attributes the performance of Fusion.
It's because we're awesome 
More seriously, even if I knew specifically what we do better ( I don't, though we do have dedicated performance people), I don't think we'd want to tell Parallels what they need to focus on.
I am guessing both products must be using hardware-assisted virtualization features of CPUs.
By default, Fusion uses the binary translation monitor for 32-bit guests, though you can tell it to use hardware-assist instead by editing the .vmx config file.
However, is it Fusion's expertise in binary translation ,compared to a counterpart of Parallels Desktop for Mac if it even has such, that helps boost the performance along with hardware-assisted features?
My understanding is that Parallels also has a binary translation monitor, though they default to the hardware-assist. One key point is that it's hardware assisted - there's still a software component, and so two hardware-assisted monitors may perform differently.